like like alike joint friendship and interest propagation in social networks shuang hong yang, bo...
TRANSCRIPT
Like like alikeJoint friendship and interest propagation
in social networks
Shuang Hong Yang, Bo Long, Alex SmolaNara Sadagopan, Zhaohui Zheng, Hongyan Zha
Georgia Institute of Technology Yahoo! Research
Outline
• Factorization models• Friendship-Interest Propagation• Experiments• Summary
Social Network Data
Data: users, connections, featuresGoal: suggest connections
x x’
y y’
e
Social Network Data
Data: users, connections, featuresGoal: model/suggest connections
x x’
y y’
e
Direct application of the Aldous-Hoover theorem.
Applications
Applicationssocial network = friendship + interests
recommend users based on friendship &
interests
recommend apps based on friendship &
interests
Social Recommendationrecommend users
based on friendship & interests
recommend apps based on friendship &
interests
• boost traffic• make the user
graph more dense
• increase user population
• stickiness
• boost traffic• increased revenue• increased user
participation• make app graph
more dense
... usually addressed by separate tools ...
Homophilyrecommend users
based on friendship & interests
recommend apps based on friendship &
interests
• users with similar interests are more likely to connect
• friends install similar applications
Highly correlated. Estimate both jointly
Model
x x’
y y’
e
v
u
s
(latent) appfeatures (latent) user
features
app install
Model
x x’
y y’
e
v
u
a
• Social interaction
• App install
Model• Social interaction
• App install
cold start latent features
bilinear features
Optimization Problem
minimize social
app
regularizer
reconstruction
Loss Function
Loss
• Much more evidence of application non-install(i.e. many more negative examples)
• Few links between vertices in friendship network (even within short graph distance)
• Generate ranking problems (link, non-link) with non-links drawn from background set
Loss
applicationrecommendation
socialrecommendation
Optimization• Nonconvex optimization problem• Large set of variables
• Stochastic gradient descenton x, v, ε for speed
• Use hashing to reducememory load, i.e.
binary hash hash
Y! Pulse
Y! Pulse Data
1.2M users, 386 items
6.1M friend connections
29M interest indications
App Recommendation
SIM: similarity based model; RLFM: regression based latent factor model (Chen&Agarwal); NLFM:
SIM&RLFM
Social recommendation
v
u
a
• Multiple relations
(user, user)(user, app)(app, advertisement)
• Users visiting several propertiesnews, mail, frontpage, social network, etc.
• Different statistical models• Latent Dirichlet Allocation for latent factors• Indian Buffet Process
v
u
a
Extensions
x x’
y y’
e
Summary
• Factorization model for users• Integration of preferences helps both
social and app recommendation• Simple stochastic gradient descent
algorithm• Hashing for memory compression• Extensible framework